Research map / Image generation / GANs
Discriminator-Cooperated Feature Map Distillation for GAN Compression
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace pixel-wise feature matching with discriminator-cooperated feature map distillation | compressed CycleGAN · horse2zebra | FID: 61.53 -> 48.24 | Helped |
| use DCD instead of OMGD | compressed CycleGAN · horse2zebra | FID: 51.92 -> 48.24 | Helped |
| use DCD instead of OMGD | compressed CycleGAN · summer2winter | FID: 73.79 -> 73.63 | Helped |
| use DCD instead of OMGD | compressed Pix2Pix · edges2shoes | FID: 25.00 -> 23.43 | Helped |
| replace DCD with FitNet pixel-wise feature matching | compressed CycleGAN · horse2zebra | FID: 48.24 -> 65.84 | Hurt |
| replace DCD with MGD pixel-wise feature matching | compressed CycleGAN · horse2zebra | FID: 48.24 -> 67.57 | Hurt |
| replace DCD with KRD cross-layer feature matching | compressed CycleGAN · horse2zebra | FID: 48.24 -> 61.53 | Hurt |
| omit perceptual loss from the full training objective (ablation) | DCD CycleGAN · horse2zebra | FID: 48.24 -> 323.63 | Hurt |
| omit DCD loss from the full training objective (ablation) | DCD CycleGAN · horse2zebra | FID: 48.24 -> 65.13 | Hurt |
| omit adversarial loss from the full training objective (ablation) | DCD CycleGAN · horse2zebra | FID: 48.24 -> 55.24 | Hurt |
| use DCD loss without perceptual or adversarial loss (ablation) | CycleGAN · horse2zebra | FID: 48.24 -> 324.27 | Hurt |
| set λdcd to 0.1 instead of 1 (ablation) | DCD CycleGAN · horse2zebra | FID: 48.24 -> 58.76 | Hurt |
Closest papers by meaning
- Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme (NeurIPS 2021 · similarity 0.94)
- Dual Discriminator Generative Adversarial Nets (NeurIPS 2017 · similarity 0.90)
- Distilling Portable Generative Adversarial Networks for Image Translation (AAAI 2020 · similarity 0.90)
- Teachers Do More Than Teach: Compressing Image-to-Image Models (CVPR 2021 · similarity 0.90)
- AutoGAN-Distiller: Searching to Compress Generative Adversarial Networks (ICML 2020 · similarity 0.90)
- GAN Compression: Efficient Architectures for Interactive Conditional GANs (CVPR 2020 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.